sql: update demo for new multi-region abstractions

[awoods187]

We currently have both cockroach demo --geo-partitioned-replicas and global. Neither exactly makes sense for 21.1. with our new abstractions.

I'm proposing a new --multi-region that shows the idealized state using movr like the previous --geo-partitioned-replicas (and deprecating the --geo-partitioned-replicas).

I'm also proposing a clean demo slate with locality set up for general docs usage or to build up demos. I currently use /cockroach demo movr --nodes 9 but perhaps without the movr?

Jira issue: CRDB-2689

[awoods187]

It's a bit late in the cycle to squeeze in changes and the team is hard at work bug fixing so we are proposing the following plan of attack:

• Deprecating --geo-partitoned-replicas from our docs but not our codebase in 21.1
• Using ./cockroach demo --nodes 9 --empty --global to start from scratch (it contains 3 nodes each in us-east1, us-west1, and europe-west1`. This can be the starting point for non-movr related demos.
• Using ./cockroach demo --movr --nodes 9 --global for demonstrating how to convert movr from a single-region database into a multi-region database. It will come with limitations including the pre 21.1 schema (e.g., compound-primary keys) but can be used to demonstrate all of the DDL as well as auto-homing inserts and locality optimized searches. This will not show latency.
• Adding in 21.1.X (e.g., backporting) a --multi-region flag with a new movr schema and showing the correct multi-region abstractions by default.

What do you think @rmloveland and @jseldess?

[awoods187]

@jordanlewis any concerns with including --global here? I think its really nice to include but we could potentially drop it if there were concerns and use ./cockroach demo --nodes 9 --empty

[andreimatei]

Consider roachprod local in this conversation. At least for development, that's a much more useful platform cause you don't lose all the state with restarts.

[jordanlewis]

--global is unmaintained and fairly brittle and non-extensible. It might work okay for this purpose, but I'm concerned that people playing with it at our behest will run into confusing issues that we won't have the bandwidth to fix. I think the demo pipeline is pretty important to keep clean, because people will likely judge our product based on it after downloading.

If the docs team is okay with "testing" --global's suitability for this project in its current state, I feel reasonably okay about using it for the docs. But it's worth noting that we don't quality-test this feature past a very basic test that ensures that the database starts up.

As for brittle, this is what we do for the latencies. It's hardcoded and not really changeable for now. Note that we wouldn't have flexibility to change the number of nodes to be more than 9 either.

func init() {
	regionToRegionToLatency = make(map[string]map[string]int)
	// Latencies collected from http://cloudping.co on 2019-09-11.
	for pair, latency := range map[regionPair]int{
		{regionA: "us-east1", regionB: "us-west1"}:     66,
		{regionA: "us-east1", regionB: "europe-west1"}: 64,
		{regionA: "us-west1", regionB: "europe-west1"}: 146,
	} {
		insertPair(pair, latency)
		insertPair(regionPair{
			regionA: pair.regionB,
			regionB: pair.regionA,
		}, latency)
	}
}

[awoods187]

I think it would be fine for us to use this behind the scenes for katacoda as we can control the number of nodes etc. I also think it would be fine to list the known limitations for docs and demos as providing a hard-coded configuration. cc @jseldess @rmloveland

Can we add some tests in the stability period now to harden this up? I'm not asking to extend the capabilities just to make 9 nodes with --global well tested enough for us in docs and to back katacoda like we do for the spatial tutorial.

[rmloveland]

If the docs team is okay with "testing" --global's suitability for this project in its current state, I feel reasonably okay about using it for the docs

I'll take a look at it. I'm not feeling super psyched about it due to old tech writer saying: "you have to build it in your basement". This is pretty much the opposite of that for a multi-region deployment. Feels like we are not giving users the real dogfood. (Also why roachprod is DQ'd - not dog food)

However, if it can do the following, it could be made to ~work, maybe?

1. start cluster, latencies are "high"
2. run some multi-region SQL where we say "make this table GLOBAL, make this table REGIONAL"
3. cockroach moves some things around
4. the "high" latencies start to "drop" in the DB Console

Will look at it.

[jseldess]

Thanks for investigating, Rich. Overall, I agree with your point in the linked ticket that setup/provisioning shoudn't be the focus of the tutorial, so if we can this to work via demo or docker, that'd be awesome.

[rmloveland]

Did some testing and wrote up my findings so far in a wiki page: https://cockroachlabs.atlassian.net/wiki/x/BQDpXg

tl;dr: I think it works? but I may be tricking myself, and it's not super easy to set up (still easier than doing everything via docker tho IMO)

[awoods187]

@ajstorm have our concerns with demo been sufficiently addressed to check some parts of this issue off?

[ajstorm]

@awoods187 I checked off the one item we got to in 21.1 above. We need your movr changes before we can make progress on the rest of the items.

[awoods187]

is the --global work all complete now? I somehow lost the tracking issue for that

[ajstorm]

I don't think it's all complete just yet. @otan can you comment on what's left?

…

On Wed, May 5, 2021 at 7:53 PM Andy Woods ***@***.***> wrote: is the --global work all complete now? I somehow lost the tracking issue for that — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#62025 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AEMXVOREM7OKWGNMS5GZ4YLTMHK7DANCNFSM4ZG7XJ6Q> .

[otan]

--global is slightly less experimental than before, and is ready to be consumed by the user with many caveats inbuilt.

[github-actions]

We have marked this issue as stale because it has been inactive for
18 months. If this issue is still relevant, removing the stale label
or adding a comment will keep it active. Otherwise, we'll close it in
10 days to keep the issue queue tidy. Thank you for your contribution
to CockroachDB!